With the increasing availability of completely sequenced genomes and technologies for measuring the expression of thousands of genes, there is now increasing optimism that understanding transcriptional network dynamics can become a predictive discipline. Just as we can routinely predict the function of novel genes by sequence homology, it would be desirable to predict the context-dependent transcriptional activity of a gene from the DNA sequence features within its regulatory (non-protein coding) region. We propose a comprehensive inter-disciplinary research program, aimed at establishing the above paradigm. Accordingly, our specific aims are to: (1) use Bayesian networks to learn causal relationships between cis-regulatory motifs and gene expression patterns; (2) use inter-species conservation to identify cis-regulatory motifs, learn their functional constraints, and model their evolution; (3) extend specific aims 1 and 2 to the study of transcription in metazoan genomes; (4) apply a high-throughput phage-display selection strategy to identify transcription factors which bind computationally predicted cis-regulatory motifs. If implemented, the proposed research program will significantly enhance our understanding of transcriptional network structure and dynamics. On a practical level, this knowledge will set the foundation for engineering of custom regulatory circuits and rational interventions to affect human disease processes.